OpenAI Disrupts Large-Scale Chain-of-Thought Distillation Campaign, Names Moonshot AI

OpenAI disclosed and blocked a large-scale extraction campaign targeting its models' hidden reasoning content (chain-of-thought), and for the first time publicly alleged that individuals tied to Chinese company Moonshot AI (maker of Kimi) were behind a core part of the operation. The attacks began on July 1; within two days OpenAI logged roughly 16000 extraction attempts from more than 4000 users. OpenAI has since banned the associated accounts and hardened its defenses. This is the first time OpenAI has publicly accused a China-linked AI lab of distilling its models, and the follow-ups from both sides and how the situation develops deserve close attention.

Confirmed

• OpenAI claims it detected and dismantled the coordinated campaign: attackers mass-manipulated conversations to make the model expose its protected reasoning content, aiming to use that chain-of-thought data to train or improve another model (i.e., "distillation")
• As relayed by @rohanpaulai, the attackers did not break any encryption; instead they copied a piece of encrypted reasoning content and had the model in another session decrypt and transcribe it
• The campaign started July 1, logging about 16000 extraction attempts from 4000+ users within two days
• OpenAI banned the accounts involved, hardened protections, and took action via Frontier Mod (a related safety/model protection mechanism)

Not yet confirmed

• Moonshot AI has not responded in the available material, and OpenAI's one-sided accusation remains unverified by the other party

Why it matters

• Distillation means using a stronger model's outputs to train your own, which violates OpenAI's terms of use; OpenAI has repeatedly warned it would ban such behavior. This first public naming of a Chinese lab could intensify US-China AI disputes over model security and data usage

2026-10-01 ~ 2026-10-01 · 6 related posts

• [source] OpenAI Accuses Chinese Lab of Distillation: 16,000 Requests in 2 Days to Extract Hidden Reasoning — rohanpaul_ai · 2026-10-01
• OpenAI Says It Caught a Major Distillation Run, Points to Moonshot AI — ShakeelHashim · 2026-10-01
• [source] OpenAI says Moonshot AI-linked individuals led campaign making 16,000 attempts to extract hidden reasoning — kimmonismus · 2026-10-01
• Follow-up link to OpenAI's disclosure of Moonshot AI-linked hidden reasoning extraction campaign — kimmonismus · 2026-10-01
• [source] OpenAI disrupts coordinated campaign to siphon protected chain-of-thought for distillation — lmoroney · 2026-10-01
• OpenAI Disrupts Coordinated Model Distillation Campaign; Redditers See Slower Chinese Releases — LocoMod · 2026-10-01